Some questions about the Delfina DSP

[Karlos]

Hi,

Despite having no real use for one I'm actually tempted to get a delfina DSP for the spare 1200. However, I'm curious about what the 56002 on the card is actually used for.

If the card is attached via the clock port I am assuming that it can't access anything on the motherboard and that basically the CPU has to perform PIO on it.

How fast is the CPU -> delfina transfer? Is it fast enough so that several streams of 16-bit audio can be written and then mixed by the DSP, or do the existing drivers do software mixing and then pass the mixed stream to the card? If it is the latter, then what role does the DSP play in normal use?

Additionally (and most importantly), can I, as a curious developer, write my own DSP code and run it on the card? What technical limitations are there with respect to the interface to the DSP (ie what restrictions other than speed and address range does the clockport impose)?

[Karlos]

@Matt_H

To be honest I was thinking about actually (mis)using the DSP for other purposes...

Either way, I don't imagine the source code for the delfina.library is available in order for it to be fixed.

[Karlos]

@Matt

Cheers, I downloaded the developer archive anyway, just to see.

The thing is, I don't really want to get the card if it is going to point blank refuse to work out of the box, which seems to be the experience some people are having by the sound of it. Has anybody ever correlated which combinations work and which don't and what sort of problem they get in the process?

[Karlos]

Aww, your'e no fun!

I actually wanted to see if I could write some realtime resonant filters for it and possibly use that as the basis of a MIDI driven sampler ;-)

It would be a nice way to get the miggy coding bug back.

[Karlos]

Mixing of audio channels is done by the DSP, of course.

Mixing of which audio channels though? If the limit is 300KB/s, two 16-bit stereo channels at 44100Hz would max the interface out. A single 16-bit stereo 44100 stream needs ~173KB/s.

Incidentally, is the clockport addressable by any of the custom chips?

[Karlos]

golem wrote:
Be warned though DSP math is uber-geek. I think you'd like it  :-D

It isn't that bad, really. It looks it when you see it formally expressed, but in the end it's all pretty much down to multiply-accumulate :-)

Last DSP stuff I did was on a Zilog 16-bit DSP.

[Karlos]

Ok, we've established that the clockport Delfina is almost useless for my original idea since it doesn't have the bandwidth to support multiple streams at 44kHz.

However, it looks like it could manage around 5 - 6 mono 22kHz 16-bit source streams before total bus saturation which means it might be able to serve as a cut down sampler. There are also various lossy 8-bit delta encodings one could apply to the source data increase that again, giving a more respectible polyphony.

The second idea I had might still be ok, which was to investigte the possiblility of a tiny softsynth that runs entirely on the DSP (it has several useful lookup tables in ROM) and simply controlled by signals from the CPU that are ultimately raised by MIDI input. The clockport has more than ample bandwidth for this.